KR102629877B1 - Method for recognizing military symbol using artificial neural network, method for training artificial neural network, and computing system performing the same - Google Patents

Abstract

Disclosed are a method for recognizing the type and rotation angle of military symbols used in establishing military operations or briefings, a learning method for an artificial neural network for this purpose, and a computing system for performing this method. According to one aspect of the present invention, an artificial neural network for military code classification for receiving a military code image and determining whether the military code image corresponds to the 1st military code to the Nth military code, and an artificial neural network for first angle judgment Neural network or artificial neural network for determining the Nth angle - Here, the artificial neural network for determining the ith angle is used to receive the image of the ith military code and determine how much the input image of the ith military code is rotated. It is a neural network - a storage module that stores the image, an acquisition module that acquires a judgment target image, and a military code that inputs the judgment target image into the artificial neural network for military code classification and determines the military code corresponding to the judgment target image. If the judgment module and the judgment target image are determined to correspond to the kth military code, the judgment target image is input to the kth angle judgment artificial neural network, and the judgment target image is compared to the kth standard military code to what extent A military code recognition system is provided that includes a rotation angle determination module that determines the rotation angle indicating whether the device is rotated.

Description

Method for recognizing military symbol using artificial neural network, method for recognizing military symbol using artificial neural network, method for training artificial neural network, and computing system performing the same}

The present invention relates to a military code recognition method using an artificial neural network, an artificial neural network learning method therefor, and a computing system that performs the same. In more detail, it relates to a method of recognizing the type and rotation angle of military symbols used for establishing military operations or briefings using a neural network, a learning method for the artificial neural network for this, and a computing system that performs this method. will be.

Recently, due to the development of machine learning, attempts are being made to automate tasks such as image recognition or classification using computer systems. In particular, attempts are being made to effectively recognize images using artificial neural networks, a type of machine learning. Additionally, there have been many performance improvements in image analysis technology recently by utilizing GPU-based high-speed computation, large amounts of data, and deep learning. Therefore, there are attempts to apply this deep learning-based image analysis technology to various fields, one of which is the field of military operations conducted by the military or military institutions.

When briefing military operations using a computer, in most cases, the person in charge of the briefing explains the operation by displaying various military codes on the operation map screen. Previously, the desired symbol was used on the GUI (Graphic User Interface), which separately collected military codes to be used. Military symbols were displayed on the map by selecting and placing them on the map screen. In addition, because the direction indicated by military symbols is important in military operations, briefing staff had to separately perform operations to rotate military symbols in a specific direction (for example, dragging to rotate a shape, etc.). Due to this inconvenient operation method, there is a tendency to avoid military operation briefings using computers in the field. In order to resolve the inconvenience of this operation method, when the person in charge draws the necessary symbol in the desired direction, it is automatically recognized and displayed on the map screen. The demand for technology is increasing.

Republic of Korea Patent Publication No. 10-2018-0080051

The technical problem to be achieved by the present invention is to determine which type of military code used in military operations the input image corresponds to and to determine at what angle it is rotated through an artificial neural network. It is to implement.

In particular, the learning efficiency and recognition of the neural network can be increased by distinguishing between the neural network for distinguishing the types of military symbols and the neural network for determining the rotation angle. Furthermore, the neural network for angle judgment can be used for each military unit. It provides a method and system that can significantly increase learning efficiency and recognition by building each code separately.

According to one aspect of the present invention, a military code classification system that receives a military code image and outputs information to determine whether the military code image corresponds to the 1st military code or the Nth military code (N is an integer of 2 or more). for artificial neural networks; and an artificial neural network for determining the first angle or an artificial neural network for determining the N-th angle - where the artificial neural network for determining the i-th angle (i is an arbitrary integer of 1<=i<=N) is the i-th group sign. It is a neural network that receives the image of the i military code and outputs information to determine how much the image of the input i military code is rotated compared to the predetermined i standard military code. A storage module, an acquisition module for acquiring a judgment target image, inputting the judgment target image into the artificial neural network for military code classification, and corresponding to the judgment target image based on the value output by the military code classification artificial neural network. If it is determined that the military code judgment module determines the military code and the judgment target image corresponds to the kth military code (k is any of the integers 1<=k<=N), the judgment target image is kth Input to an artificial neural network for angle judgment, and determine a rotation angle indicating how much the judgment target image is rotated compared to the kth standard military code based on the value output by the artificial neural network for angle judgment. A military code recognition system including a rotation angle determination module is provided.

In one embodiment, each of the 1st military unit code through the N military unit code is an infantry division, an infantry brigade, an infantry battalion, an infantry company, an infantry platoon, an infantry squad, a reconnaissance battalion, a reconnaissance company, a reconnaissance platoon, an artillery brigade, a gun barrel artillery battalion, Gun artillery company, gun barrel group, self-propelled artillery battalion, self-propelled artillery company, self-propelled artillery group, armored brigade, tank battalion, tank company, tank platoon, tank squad, mechanized infantry platoon, mechanized infantry squad, mechanized infantry division, mechanized infantry brigade, mechanized infantry Battalion, mechanized infantry company, engineer brigade, engineer battalion, engineer company, engineer platoon, engineer squad, information and communications brigade, information and communications battalion, information and communications company, communications platoon, communications squad, aviation brigade, aviation battalion, aviation company, chemical, biological and biological brigade. , a chemical, biological and radiological battalion, a chemical, biological and radiological defense company, a corps, an attack helicopter battalion, a medical unit, a medical battalion, an air defense platoon, a mortar platoon, a large node communication station, a small node communication station, a unit communication station and a relay station. You can.

In one embodiment, the artificial neural network for military code classification includes: a first network that detects the widest closed curve having the shape of a square or circle in an input military code image; A second network that classifies a combination of shapes included in the inner region of the closed curve detected by the first network among the military code images, and a figure included in the outer region of the closed curve detected by the first network among the military code images. A third network that classifies the combination and outputs information to determine which of the 1st military code to the Nth military code the military code image corresponds to based on the result values output from the second network and the third network. It may include a fourth network.

In one embodiment, a first learning module for learning the artificial neural network for military code classification; and a second learning module that learns the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle, wherein the first learning module is configured to select any one of the first group code to the N-th group code. An artificial neural network for military code classification is learned by inputting a plurality of images for training a military code classification model tagged with information expressing whether it corresponds to a classification, and image augmentation for each of the multiple military code classification model learning images ( image augmentation) is performed to generate a plurality of images for additional training of the military code classification model, and the generated images for additional training of the military code classification model are input to further learn the artificial neural network for military code classification - here, the military code classification Each model additional training image is tagged with the label of the original image before image augmentation, and the second learning module is configured for each of the plurality of rotation angle classification model training images - where the rotation angle classification model training image is, Contains a plurality of military code classification model training images and the plurality of military code classification model additional training images - Rotating the rotation angle classification model training image by a predetermined unit angle and plurality of images corresponding to the rotation angle classification model training image Create a rotated image - each of the plurality of rotated images is tagged with a rotation angle - and input the plurality of rotated images to learn an artificial neural network for determining the angle corresponding to the image for training the rotation angle classification model. can do.

In one embodiment, the judgment target image may be a handwriting image drawn by a user using a predetermined input device.

In one embodiment, the military code recognition system includes a map display module that displays a map image on a predetermined display device, a UI providing module that provides a drawing UI for drawing the handwritten image, and the k standard military code. It may further include a creation module that generates an object image rotated by the rotation angle, and an object display module that displays the object image at a specific point on the map image selected by the user.

In one embodiment, the acquisition module periodically acquires the judgment target image while the user is drawing on the drawing UI, or acquires the judgment target image whenever a line drawn by the user on the drawing UI is disconnected. can do.

According to another aspect of the present invention, a military code that receives a military code image and outputs information to determine whether the military code image corresponds to the 1st military code to the Nth military code (N is an integer of 2 or more). An artificial neural network for classification and an artificial neural network for judging the first angle or an artificial neural network for judging the Nth angle - where, the artificial neural network for judging the ith angle (i is a random integer with 1<=i<=N) , A neural network that corresponds to the i military code and receives the image of the i military code and outputs information to determine how much the input image of the i military code is rotated compared to the predetermined i standard military code. A storage module for storing the -, a first learning module for learning the artificial neural network for military code classification, and a second learning module for learning the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle. It includes, but the first learning module inputs a plurality of images for training a military code classification model tagged with information representing which category it corresponds to from the 1st military code to the Nth military code, and creates an artificial neural network for classifying the military code. Learn the network, perform image augmentation on each of the multiple military code classification model training images to generate multiple images for additional training of the military code classification model, and create images for additional training of the military code classification model. Input to further learn the artificial neural network for military code classification - Here, each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation, and the second learning module has a plurality of For each rotation angle classification model training image - where the rotation angle classification model training image includes the plurality of military code classification model training images and the plurality of military code classification model additional training images -, the rotation angle classification An image for model training is rotated by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training image - each of the plurality of rotated images is tagged with a rotation angle - and the plurality of rotated images are tagged with a rotation angle. A neural network learning system for military code recognition is provided that inputs an image and learns an artificial neural network for determining angles corresponding to the image for training the rotation angle classification model.

According to another aspect of the present invention, a computing system receives a military code image and provides information to determine which of the 1st military code to the Nth military code (N is an integer of 2 or more) that the military code image corresponds to. An artificial neural network for output military code classification and an artificial neural network for determining the first angle or an artificial neural network for determining the N-th angle - where, an artificial neural network for determining the i-th angle (i is an arbitrary 1<=i<=N (is an integer of) corresponds to the i military code, receives the image of the i military code, and provides information to determine how much the image of the input i military code is rotated compared to the predetermined i standard military code. A step of storing the output neural network - a step of the computing system acquiring a judgment target image; The computing system inputs the judgment target image into the artificial neural network for military code classification, and determines the military code corresponding to the judgment target image based on the value output by the artificial neural network for military code classification. And when it is determined that the judgment target image corresponds to the kth military code (k is any one of the integers 1<=k<=N), the computing system uses a neural network for determining the kth angle of the judgment target image. military code recognition, including the step of inputting a rotation angle indicating how much the judgment target image is rotated compared to the kth standard military code based on the value output by the neural network for determining the kth angle. A method is provided.

In one embodiment, the computing system, learning the artificial neural network for military code classification, and the computing system, learning the artificial neural network for determining the first angle to the artificial neural network for determining the Nth angle. Further comprising the step of learning the artificial neural network for military code classification, the step of learning a plurality of military code classification model training images tagged with information representing which classification among the 1st military code to the Nth military code corresponds to learning the artificial neural network for military code classification by inputting; Image augmentation is performed on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and the generated military code classification model additional training images are input to the military code classification model. A step of further learning an artificial neural network for code classification - where each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation - and an artificial neural network for determining the first angle. The step of learning the network or the artificial neural network for determining the N-th angle includes, for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training image is the plurality of military code classification model training images and the Contains a plurality of military code classification model additional training images - Rotating the rotation angle classification model training image by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training images - The plurality Each of the rotated images is tagged with a rotation angle, and may include the step of inputting the plurality of rotated images and learning an artificial neural network for determining angles corresponding to the image for training the rotation angle classification model.

In one embodiment, the military code recognition method includes, by the computing system, displaying a map image on a predetermined display device, providing a drawing UI for drawing the handwriting image by the computing system, the computing The system may further include generating an object image in which the kth standard military code is rotated by the rotation angle, and the computing system displaying the object image at a specific point on the map image selected by the user. You can.

In one embodiment, the step of acquiring the judgment target image includes periodically acquiring the judgment target image while the user is drawing on the drawing UI or whenever a line drawn by the user on the drawing UI is disconnected. It may include obtaining the judgment target image.

According to another aspect of the present invention, a computing system receives a military code image and provides information to determine which of the 1st military code to the Nth military code (N is an integer of 2 or more) that the military code image corresponds to. The output artificial neural network for military code classification and the computing system include an artificial neural network for determining the first angle or an artificial neural network for determining the Nth angle - where, an artificial neural network for determining the ith angle (i is 1<=i <=N is an arbitrary integer) corresponds to the i military code, receives the image of the i military code and determines how much the input image of the i military code is rotated compared to the predetermined i standard military code. A step of storing a neural network that outputs information for judgment; learning, by the computing system, the artificial neural network for military code classification; and a step of, by the computing system, learning the artificial neural network for determining the first angle or the artificial neural network for determining the Nth angle, wherein the step of learning the artificial neural network for classifying the military code includes: Learning an artificial neural network for military code classification by inputting a plurality of images for training a military code classification model tagged with information representing which classification among the codes or Nth military codes it corresponds to; And performing image augmentation on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and inputting the generated military code classification model additional training images. Further learning an artificial neural network for military code classification - wherein each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation, wherein the first The step of learning the artificial neural network for angle determination or the artificial neural network for determining the Nth angle includes, for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training image is the plurality of military code classifications. Contains images for model training and images for additional training of the military code classification model - Rotate the rotation angle classification model training image by a predetermined unit angle to create a plurality of rotated images corresponding to the rotation angle classification model training images. Generating - each of the plurality of rotated images is tagged with a rotation angle - and inputting the plurality of rotated images to learn an artificial neural network for determining angles corresponding to the images for training the rotation angle classification model. A neural network learning method for military code recognition is provided.

According to another aspect of the present invention, a computer program installed in a data processing device and recorded on a medium for performing the above-described method is provided.

According to another aspect of the present invention, a computer-readable recording medium on which a computer program for performing the above-described method is recorded is provided.

According to another aspect of the present invention, a computing system comprising: a processor; and a memory storing a computer program, wherein the computer program, when executed by the processor, causes the computing system to perform the method described above.

According to the technical idea of the present invention, a method and system that uses an artificial neural network to determine which type of military code a particular image corresponds to among various military codes used in military operations and to determine at what angle it is rotated. can be provided.

In particular, the learning efficiency and recognition of the neural network can be increased by distinguishing between the neural network for distinguishing the types of military symbols and the neural network for determining the rotation angle. Furthermore, the neural network for angle judgment can be used for each military unit. By constructing each code separately, it is possible to provide a method and system that can significantly increase learning efficiency and recognition.

In order to more fully understand the drawings cited in the detailed description of the present invention, a brief description of each drawing is provided.
Figure 1 is a diagram schematically showing an environment in which a military code recognition method using an artificial neural network and an artificial neural network learning method for military code recognition are performed according to the technical idea of the present invention.
Figures 2a and 2b are diagrams showing 54 types of standard military codes according to an embodiment of the present invention. Figure 2c is a diagram for explaining the rotation angle of the military symbol.
Figure 3 is a diagram showing the specific configuration of an artificial neural network according to an embodiment of the present invention.
Figure 4 is a diagram showing the detailed structure of an artificial neural network for military code classification according to an embodiment of the present invention.
Figure 5 is a block diagram showing the configuration of a neural network learning system according to an embodiment of the present invention.
Figure 6 is a flowchart illustrating an example of a process in which a neural network learning system learns an artificial neural network according to an embodiment of the present invention.
Figure 7 is a block showing the configuration of a military code recognition system according to an embodiment of the present invention.
Figure 8 is a diagram illustrating an example of a military code recognition method according to an embodiment of the present invention.

Since the present invention can be modified in various ways and can have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all transformations, equivalents, and substitutes included in the spirit and technical scope of the present invention. In describing the present invention, if it is determined that a detailed description of related known technologies may obscure the gist of the present invention, the detailed description will be omitted.

Terms such as first, second, etc. may be used to describe various components, but the components should not be limited by the terms. Terms such as first, second, etc. do not indicate a particular order and are used only for the purpose of distinguishing one component from another component.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this specification, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are intended to indicate the presence of one or more other It should be understood that this does not exclude in advance the presence or addition of features, numbers, steps, operations, components, parts, or combinations thereof.

Additionally, in this specification, when one component 'transmits' data to another component, the component may transmit the data directly to the other component, or through at least one other component. This means that the data can be transmitted to the other components. Conversely, when one component 'directly transmits' data to another component, it means that the data is transmitted from the component to the other component without going through the other component.

Hereinafter, the present invention will be described in detail focusing on embodiments of the present invention with reference to the attached drawings. The same reference numerals in each drawing indicate the same member.

Figure 1 shows a military code recognition method using an artificial neural network (hereinafter referred to as 'military code recognition method') and an artificial neural network learning method for military code recognition (hereinafter referred to as 'artificial neural network learning method') according to the technical idea of the present invention. This is a diagram schematically showing the environment in which this is performed.

Referring to FIG. 1, the artificial neural network learning method according to an embodiment of the present invention can be performed by the neural network learning system 200 for military code recognition, and the military code recognition method according to an embodiment of the present invention Can be performed by the military code recognition system 200.

The military code recognition system 100 uses the artificial neural network 300 to recognize N military codes (i.e., 1st military code to Nth military code) with predetermined judgment target images (N is an integer of 2 or more). ), it is possible to determine which type it belongs to, and how much it has been rotated compared to the standard military code of the corresponding type.

The artificial neural network 300 may have been previously trained by the neural network learning system 200 for military code recognition.

The military code recognized by the military code recognition system 100 may be a standardization of various symbols used in operational maps, and may mean a special symbol displayed on a map or situation map to indicate unit deployment or situation in the military. It may be possible. In addition, the military code recognized by the military code recognition system 100 may be a symbol representing a military organization used when establishing an operation in the military or briefing an established operation. The military code may be a common standard code, or it may be a code used only by a specific country or organization. For example, the military symbol may be a NATO joint military symbol, etc.

Meanwhile, a standard object representing each of the N military codes (hereinafter referred to as 'standard military code') may be predefined. Figures 2a and 2b are diagrams showing 54 types of standard military codes according to an embodiment of the present invention. That is, in the examples of FIGS. 2A and 2B, N is 54.

As shown in Figures 2a and 2b, each of the 54 types of military codes is infantry division, infantry brigade, infantry battalion, infantry company, infantry platoon, infantry squad, reconnaissance battalion, reconnaissance company, reconnaissance platoon, artillery brigade, and artillery battalion. , gun barrel artillery company, gun barrel group, self-propelled artillery battalion, self-propelled artillery company, self-propelled artillery group, armored brigade, tank battalion, tank company, tank platoon, tank squad, mechanized infantry platoon, mechanized infantry squad, mechanized infantry division, mechanized infantry brigade, mechanized Infantry battalion, mechanized infantry company, engineer brigade, engineer battalion, engineer company, engineer platoon, engineer squad, information and communications brigade, information and communications battalion, information and communications company, communications platoon, communications squad, aviation brigade, aviation battalion, aviation company, chemical and biological defense It may be a symbol representing any one of the following: brigade, chemical, biological and radiological battalion, chemical, biological and radiological defense company, corps, attack helicopter battalion, medical corps, medical battalion, air defense platoon, mortar platoon, large node communication center, small node communication center, unit communication center and relay station.

In one embodiment, the military code recognition system 100 acquires a judgment target image drawn through a predetermined input device 20, and determines what type of military code the obtained judgment target image corresponds to. , it is possible to determine at what angle the image is rotated compared to the corresponding standard military code.

FIG. 2C is a diagram illustrating an example of an image to be recognized by the military code recognition system 100. For example, if the judgment target image is the same as the portion indicated by a thick line in the image as shown in FIG. 2C, the military code recognition system 100 determines that the judgment target image in FIG. 2C is a military code representing 'infantry division'. It can be judged that it is rotated clockwise by θ compared to the standard military symbol representing 'infantry division'.

The military code recognition system 100 may express the rotated angle as a single number, but depending on the embodiment, the military code recognition system 100 may express the degree of rotation as an angle section or a specific value representing the corresponding angle section. It can also be displayed. For example, the military code recognition system 100 can distinguish rotation angles in units of 10°.

The military code recognition system 100 can rotate the standard military code corresponding to the recognized military code by the recognized angle and display it on a predetermined display device 30. Meanwhile, in one embodiment, a map for military operations may be displayed on the display device 30, and the military symbol recognition system 100 displays a rotated standard military symbol by overlaying it on the displayed map for military operations. can do.

The input device 20 may be, for example, a mouse, keyboard, stylus pen, tablet, touch panel, or touch screen.

The display device 30 may be, for example, an LCD, an LED display device, an OLED display device, or a CRT display device.

The input device 20 and/or the display device 30 may be a peripheral device of the military code recognition system 100, but is not necessarily limited thereto, and is connected to the military code recognition system 100 through a network. It may be a peripheral device of an external terminal (not shown).

The neural network learning system 200 for military code recognition and/or the military code recognition system 100 may be a computing system that is a data processing device with computing capabilities for implementing the technical idea of the present invention, and is generally a network It may include computing devices such as personal computers and mobile terminals as well as servers, which are data processing devices that can be accessed by clients.

The neural network learning system 200 for military code recognition and/or the military code recognition system 100 may be implemented as any one physical device, but if necessary, a plurality of physical devices may be organically combined to form the present invention. An average expert in the technical field of the present invention can easily deduce that the neural network learning system 200 for military code recognition and/or the military code recognition system 100 can be implemented according to the technical idea.

As shown in FIG. 1, the neural network learning system 200 for military code recognition and/or the military code recognition system 100 may be implemented in the form of a subsystem of a predetermined mother system 10. It may be possible. The parent system 10 may be a server. The server 10 refers to a data processing device with computing power to implement the technical idea of the present invention, and is generally a data processing device that can be accessed by a client through a network, as well as providing a specific service such as a personal computer, mobile terminal, etc. An average expert in the technical field of the present invention can easily deduce that any device capable of performing the above-described tasks can be defined as a server.

Alternatively, depending on the embodiment, the neural network learning system 200 for military code recognition and the military code recognition system 100 may be implemented in a separate form.

Meanwhile, the artificial neural network 300 may be composed of a plurality of artificial neural networks, and FIG. 3 is a diagram showing the specific configuration of the artificial neural network 300 according to an embodiment of the present invention.

Referring to FIG. 3, the artificial neural network 300 may include an artificial neural network 310 for military code classification and a neural network 320 for determining N angles.

The artificial neural network 310 for military code classification may receive a military code image and output information for determining whether the military code image corresponds to the 1st military code or the Nth military code. For example, the artificial neural network 310 for military code classification may determine the probability that the military code image corresponds to the first military code, the probability that the military code corresponds to the second military code,... , the probability of corresponding to the Nth military code can be output.

The N angle judgment neural networks 320 (320-1 to 320-N) may correspond one-to-one to the N military codes. That is, the ith angle judgment neural network 320-i may correspond to the ith military code (i is an arbitrary integer where 1<=i<=N).

In addition, the artificial neural network 320-i for determining the i-th angle is used to receive an image of the i-th military code and determine how much the input image of the i-th military code is rotated compared to the i standard military code. Information can be printed. For example, the artificial neural network 320-i for determining the i-th angle may include a probability that the input image of the i-th military code is not rotated compared to the i-th standard military code, a probability that it is rotated by r°, Probability of being rotated by 2r°, … , the probability of being rotated by (n-1)×r° can be output (r is a positive real number, n = 360/r).

In this specification, the artificial neural network is a neural network artificially constructed based on the operating principles of human neurons, includes a multi-layer perceptron model, and can refer to a set of information expressing a series of design details that define the artificial neural network. there is. The artificial neural network 310 for military code classification and the neural network 320 for determining the N angles may be supervised learning models that are learned using learning data labeled with correct answers. In one embodiment, the artificial neural network 310 for military code classification and the neural network 320 for determining the N angles may be a convolutional neural network or may include a convolutional neural network.

As is well known, the convolutional neural network may include an input layer, a plurality of hidden layers, and an output layer. Each of the plurality of hidden layers may include a convolution layer and a pooling layer (or sub-sampling layer). A convolutional neural network can be defined by functions, filters, strides, weight factors, etc. to define each of these layers. Additionally, the output layer may be defined as a fully connected feedforward layer. The design details of each layer that makes up a convolutional neural network are widely known. For example, known functions may be used for the number of layers to be included in a plurality of layers, a convolution function, a pooling function, and an activation function for defining the plurality of layers. In order to implement the technical idea of the present invention, Separately defined functions may also be used.

N military codes may have common morphological characteristics, and the artificial neural network 310 for military code classification can be designed with a structure that can well capture these common morphological characteristics of military codes, Figure 4 is a diagram showing the detailed structure of the artificial neural network 310 for military code classification according to an embodiment of the present invention.

Referring to FIG. 4, according to an embodiment of the present invention, the artificial neural network 310 for military code classification includes a first network 311, a second network 312, a third network 313, and a fourth network. It may include a network 314.

The first network 311 may be an artificial neural network for detecting the widest closed curve in the shape of a square or circle in the input military code image.

The second network 312 may be an artificial neural network for classifying a combination of shapes included in the inner region of the closed curve detected by the first network 311 among the military code images.

The third network 313 may be an artificial neural network for classifying a combination of shapes included in an area outside the closed curve detected by the first network among the military code images.

The fourth network 314 determines which of the first to Nth military codes the military code image corresponds to based on the result values output from the second network 312 and the third network 313. It may be an artificial neural network that outputs information to do this.

Figure 5 is a block diagram showing the configuration of a neural network learning system 200 according to an embodiment of the present invention.

The neural network learning system 200 for military code recognition and the military code recognition system 100 may refer to a logical configuration equipped with hardware resources and/or software necessary to implement the technical idea of the present invention. It does not necessarily mean one physical component or one device. In other words, the neural network learning system for military code recognition 200 and the military code recognition system 100 may mean a logical combination of hardware and/or software provided to implement the technical idea of the present invention, If necessary, it may be implemented as a set of logical components to implement the technical idea of the present invention by being installed in devices separated from each other and performing each function. In addition, the neural network learning system 200 for military code recognition and the military code recognition system 100 may refer to a set of components implemented separately for each function or role to implement the technical idea of the present invention. . Each component of the neural network learning system for military code recognition 200 and the military code recognition system 100 may be located in different physical devices or may be located in the same physical device. In addition, depending on the implementation example, the combination of software and/or hardware constituting each of the components of the military code recognition neural network learning system 200 and the military code recognition system 100 is also located in different physical devices, Components located in different physical devices may be organically combined to implement each of the modules.

Additionally, in this specification, a module may mean a functional and structural combination of hardware for carrying out the technical idea of the present invention and software for driving the hardware. For example, the module may mean a logical unit of a predetermined code and a hardware resource for executing the predetermined code, and does not necessarily mean a physically connected code or a single type of hardware. can be easily inferred by an average expert in the technical field of the present invention.

Referring to FIG. 5, the neural network learning system 200 for military code recognition may include a storage module 210, a first learning module 220, and a second learning module 230. Depending on the embodiment of the present invention, some of the above-described components may not necessarily correspond to components essential for implementation of the present invention, and depending on the embodiment, neural network learning for military code recognition Of course, system 200 may include more components. For example, the neural network learning system 200 for military code recognition includes a communication module (not shown) for communicating with an external device, and a system for controlling the components and resources of the neural network learning system 200 for military code recognition. A control module (not shown) may be further included.

The storage module 210 can store the artificial neural network 300 to be learned, that is, the artificial neural network 310 for military code classification, and the artificial neural network 320 for determining the N angles. In addition, the storage module 210 can further store data to be used for learning of the artificial neural network 30 (e.g., a plurality of learning images tagged with the type of military code, a plurality of learning images tagged with a rotation angle). You can.

The first learning module 220 can learn the artificial neural network 310 for military code classification, and the second learning module 230 can learn the artificial neural network 320 for determining the N angles. You can.

FIG. 6 is a flowchart showing an example of a process in which the first learning module 220 and the second learning module 230 learn the artificial neural network 300.

Steps S100 to S130 of FIG. 6 represent a process in which the artificial neural network 310 for military code classification is learned by the first learning module 220, and steps S200 to S220 of FIG. 6 represent the process of learning the second learning module ( 230) shows the process of learning the N artificial neural networks 320 for determining angles.

Referring to FIG. 6, the first learning module 220 can learn the artificial neural network 310 for military code classification using a plurality of images for learning a military code classification model (S100).

The plurality of military code classification model learning images may be pre-stored in the storage module 210, and each of the plurality of military code classification model learning images includes a corresponding image in any of the 1st military code to the Nth military code. Information expressing whether it corresponds to may be tagged in advance.

First, the first learning module 220 can learn the artificial neural network 310 for military code classification by inputting an image for training a military code classification model into the artificial neural network 310 for military code classification (S110). .

Additionally, the first learning module 220 may convert an image for training a military code classification model and generate a plurality of additional training images corresponding to the image.

Additional training images corresponding to the military code classification model training image may be tagged with the same label as the original image (i.e., the military code classification model training image).

In one embodiment, the first learning module 220 may perform image augmentation on an image for training a military code classification model and generate a plurality of additional training images for a military code classification model corresponding to the image (S120 ). Image augmentation refers to a technique of creating additional images for learning by transforming the shape by adding artificial changes to the original image. It can be implemented by resizing, scale conversion, flipping, shearing conversion, rotation, adding noise, or a combination of these. there is.

In one embodiment, the first learning module 220 may perform affine transformation as one of the image augmentation methods to generate images for additional training of a plurality of military code classification models. An affine transformation is a transformation in which affine geometric properties are preserved and may include translation, scaling, shear, rotation, etc.

Depending on the embodiment, it goes without saying that various image transformation methods in addition to affine transformation can be applied to generate images for additional learning.

Thereafter, the first learning module 220 inputs the generated images for additional training of the plurality of military code classification models into the artificial neural network for military code classification 310 to further learn the artificial neural network 310 for military code classification. You can do it (S110).

Steps S110 to S130 may be performed on each of the plurality of military code classification model training images (S100).

The second learning module 230 can learn the artificial neural network 320 for determining the plurality of rotation angles using images for learning a plurality of rotation angle classification models (S200).

The plurality of rotation angle classification model training images may include the plurality of military code classification model training images, and may further include the plurality of military code classification model additional training images. That is, the plurality of rotation angle classification module learning images include not only the plurality of military code classification model learning images stored in the storage module 210, but also the plurality of military code classification model additional learning images newly created by converting them. More may be included.

The second learning module 230 rotates the rotation angle classification model training image by a predetermined unit angle (for example, 5° or 10°, etc.) to perform a plurality of rotations corresponding to the rotation angle classification module training image. images can be created. For example, the second learning module 230 may generate 35 rotated images by rotating the rotation angle classification model training image by 10° (S210). At this time, each of the plurality of rotated images may be tagged with a rotation angle. For example, when the second learning module 230 generates a rotated image by rotating the original image by 20°, the rotation angle of the original image + 20° may be tagged in the generated rotated image. Depending on the embodiment, the original image may be considered to be in an unrotated state (i.e., the rotation angle is 0°).

Thereafter, the second learning module 230 inputs the rotation angle classification model training image and a plurality of rotated images corresponding to the rotation angle classification model training image to determine the angle corresponding to the rotation angle classification model training image. A neural network can be learned (S220). In more detail, the rotation angle classification model training image is one of the plurality of military code classification model training images or the plurality of military code classification model additional training images, and these images include the corresponding military code classification. Since they are tagged, the second learning module 230 connects the rotation angle classification model learning image and the rotation angle classification model to an artificial neural network for angle determination corresponding to the classification of the military code tagged in the rotation angle classification model learning image. You can learn an artificial neural network for determining the angle by inputting rotated images corresponding to the training image.

Steps S210 to S220 may be performed on each of the plurality of rotation angle classification model training images (S200).

Figure 7 is a block diagram showing the configuration of the military code recognition system 100 according to an embodiment of the present invention.

Referring to FIG. 7, the military code recognition system 100 may include a storage module 110, an acquisition module 120, a military code determination module 130, and a rotation angle determination module 140. In addition, depending on the embodiment, the military code recognition system 100 may further include a map display module 150, a UI provision module 160, a creation module 170, and an object display module 180. Depending on the embodiment of the present invention, some of the above-described components may not necessarily correspond to components essential for implementation of the present invention, and depending on the embodiment, neural network learning for military code recognition Of course, system 200 may include more components. For example, the neural network learning system 200 for military code recognition includes a communication module (not shown) for communicating with an external device, and a system for controlling the components and resources of the neural network learning system 200 for military code recognition. A control module (not shown) may be further included.

Additionally, depending on the embodiment, the military code recognition system 100 may be implemented in an integrated form with the neural network learning system 200 for military code recognition. In this case, the military code recognition system 100 may be implemented as the above-described system. It may further include the first learning module 220 and the second learning module 230.

The storage module 110 can store a pre-learned artificial neural network 310 for military code classification and an artificial neural network for determining the first angle to an artificial neural network for determining the N angle (320-1 to 320-N). there is. As described above, the artificial neural network 310 for military code classification is a neural network that receives a military code image and outputs information to determine whether the military code image corresponds to the 1st military code to the Nth military code. It is a network, and the artificial neural network for determining the i angle corresponding to the i military code (i is a random integer of 1<=i<=N) receives the image of the i military code and determines the i military code of the input i military code. It is a neural network that outputs information to determine how much the image is rotated compared to the predetermined i standard military code.

The acquisition module 120 may acquire a predetermined judgment target image. The judgment target image is an image to be recognized by the military code recognition system 100. In one embodiment, the judgment target image may be a handwriting image drawn by a user using a predetermined input device 20, but is not limited thereto.

The military code determination module 110 inputs the judgment target image into the artificial neural network 310 for military code classification, and bases the judgment target on the value output by the artificial neural network 310 for military code classification. The military code corresponding to the image can be determined.

For example, the military code judgment module 110 may classify the judgment target image as an infantry division, an infantry brigade, an infantry battalion, an infantry company, an infantry platoon, an infantry squad, a search battalion, a search company, a search platoon, an artillery brigade, and a gun barrel artillery battalion. , gun barrel artillery company, gun barrel group, self-propelled artillery battalion, self-propelled artillery company, self-propelled artillery group, armored brigade, tank battalion, tank company, tank platoon, tank squad, mechanized infantry platoon, mechanized infantry squad, mechanized infantry division, mechanized infantry brigade, mechanized Infantry battalion, mechanized infantry company, engineer brigade, engineer battalion, engineer company, engineer platoon, engineer squad, information and communications brigade, information and communications battalion, information and communications company, communications platoon, communications squad, aviation brigade, aviation battalion, aviation company, chemical and biological defense It can be judged as one of the following: brigade, chemical, biological and radiological battalion, chemical, biological and radiological defense company, corps, attack helicopter battalion, medical corps, medical battalion, air defense platoon, mortar platoon, large node communication center, small node communication center, unit communication center and relay station.

When the rotation angle determination module 120 determines that the judgment target image corresponds to the kth military code (k is any of the integers 1<=k<=N), the rotation angle determination module 120 determines the judgment target image as the kth angle. input to the artificial neural network for determining the kth angle, and based on the value output by the artificial neural network for determining the kth angle, the rotation angle indicating the degree to which the judgment target image is rotated compared to the kth standard military code can be determined. . The 1st standard military code to the Nth standard military code can all have a vertical reference axis.

For example, if the military code determination module 110 determines that the judgment target image is a military code representing 'infantry brigade', the rotation angle determination module 120 determines the N angle judgment neural network 320 ), the judgment target image can be input to the artificial neural network for angle judgment corresponding to the 'infantry brigade', and the judgment target is based on the value output by the artificial neural network for angle judgment corresponding to the 'infantry brigade'. The rotation angle indicating how much the image is rotated relative to the standard military symbol corresponding to 'infantry brigade' (for example, the first symbol in the second row of FIG. 2A) can be determined.

The military code recognition system 100 according to an embodiment of the present invention can be used when establishing or briefing military operations. To this end, the military code recognition system 100 may include a map display module 150, a UI provision module 160, a creation module 170, and an object display module 180.

The map display module 150 can display a map image on a predetermined display device 30.

The map image may be an image of a map used in military operations, and may be stored in the storage module 110 in advance.

Map data (for example, GIS data, etc.) may be pre-stored in the storage module 110, and the map display module 150 generates a map image using the map data and then generates a map image using the display device ( 30) can be displayed. Depending on the embodiment, the map display module 150 may provide functions to enlarge, reduce, and move the map.

The UI providing module 160 may provide a drawing UI for drawing a handwritten image.

In one embodiment, the user can select a specific point on the map image through, for example, the input device 20 or another input device such as a pointer, and the UI providing module 160 allows the user to select the map image. By selecting a specific point on the image, a drawing UI (User Interface) for drawing a handwritten image can be displayed. The drawing UI may be displayed overlaid on the map image, and depending on the embodiment, may be displayed next to the map image. Alternatively, the drawing UI may not be displayed visually.

The user can draw an image on the drawing UI through the input device 20, and the acquisition module 120 can acquire a handwriting image drawn by the user on the drawing UI as the judgment target image.

Depending on the embodiment, the acquisition module 120 periodically acquires the judgment target image while the user is drawing on the drawing UI, or the judgment target image whenever a line drawn by the user on the drawing UI is disconnected. can be obtained. Accordingly, the judgment target image may be continuously updated while the user is completing the handwritten image. In addition, each time the judgment target image is updated, the military code judgment module 110 and the rotation angle judgment module 120 determine what type of military code the corresponding handwriting image represents for the handwriting images drawn to date, and what type of military code the corresponding handwriting image represents. It can be determined whether the device has been rotated by the required rotation angle.

Meanwhile, when the military code determination module 110 determines that the judgment target image is the kth military code among the 1st to Nth military codes, and the rotation angle determination module 120 determines the rotation angle, the generation The module 170 may generate an object image obtained by rotating the kth standard military code by the rotation angle.

The object display module 180 may display the object image at the specific point on the map image (that is, a specific point on the map selected by the user).

Figure 8 is a diagram illustrating an example of a military code recognition method according to an embodiment of the present invention.

Referring to FIG. 8, the military code recognition system 100 can display a map image on a predetermined display device 30 (S300).

Additionally, the military code recognition system 100 can provide a drawing UI for drawing a handwriting image (S310). In one embodiment, when a user selects a specific point on the map image, the military code recognition system 100 may provide a drawing UI.

The user can draw a judgment target image on the drawing UI provided with a predetermined input device 20, and the military code recognition system 100 can acquire the judgment target image drawn by the user (S320).

Meanwhile, the military code recognition system 100 inputs the judgment target image into the artificial neural network for military code classification 310, and makes the decision based on the value output by the artificial neural network for military code classification 310. The military code corresponding to the target image can be determined (S330).

In addition, when the military code recognition system 100 determines that the judgment target image corresponds to the kth military code (k is any of the integers 1<=k<=N), the judgment target image is converted to the kth angle. It is input to the neural network for judgment, and based on the value output by the neural network for judgment of the kth angle, the rotation angle indicating how much the judgment target image is rotated compared to the kth standard military code can be determined ( S340).

Thereafter, the military code recognition system 100 generates an object image by rotating the kth standard military code by the rotation angle (S350), and selects the object image at the specific point on the map image (for example, when the user A specific point on the selected map can be displayed (S360).

Meanwhile, depending on the implementation example, the military code recognition system 100 and the neural network learning system for military code recognition 200 may include a processor and a memory that stores a program executed by the processor. The processor may include a single core CPU or a multi-core CPU. The memory may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. Access to memory by processors and other components may be controlled by a memory controller.

Meanwhile, the method according to the embodiment of the present invention can be implemented in the form of computer-readable program instructions and stored in a computer-readable recording medium, and the control program and target program according to the embodiment of the present invention can also be stored on a computer. It may be stored on a readable recording medium. Computer-readable recording media include all types of recording devices that store data that can be read by a computer system.

Program instructions recorded on the recording medium may be those specifically designed and configured for the present invention, or may be known and available to those skilled in the software field.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and floptical disks. Includes magneto-optical media such as ROM, RAM, flash memory, and other hardware devices specifically configured to store and execute program instructions. Additionally, computer-readable recording media can be distributed across computer systems connected to a network, so that computer-readable code can be stored and executed in a distributed manner.

Examples of program instructions include not only machine language code such as that created by a compiler, but also high-level language code that can be executed by a device that electronically processes information using an interpreter, for example, a computer.

The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The description of the present invention described above is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential features of the present invention. will be. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as unitary may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention. .

Claims (18)

An artificial neural network for military code classification that receives a military code image and outputs information to determine whether the military code image corresponds to a 1st military code or an Nth military code (N is an integer of 2 or more); and
Artificial neural network for determining the first angle or artificial neural network for determining the N-th angle - Here, the artificial neural network for determining the i-th angle (i is an arbitrary integer with 1<=i<=N) is connected to the i-th group sign. It is a neural network that receives the image of the i military code and outputs information to determine how much the input image of the i standard military code is rotated compared to the predetermined i standard military code. module;
An acquisition module that acquires a judgment target image;
A military code determination module that inputs the judgment target image into the artificial neural network for military code classification and determines a military code corresponding to the judgment target image based on a value output by the artificial neural network for military code classification; and
If it is determined that the judgment target image corresponds to the kth military code (k is any of the integers 1<=k<=N), the judgment target image is input into the kth angle judgment artificial neural network, and A military code recognition system including a rotation angle determination module that determines a rotation angle indicating the degree to which the judgment target image is rotated compared to the k standard military code based on the value output by the artificial neural network for determining the kth angle. .
The method of claim 1, wherein each of the 1st military code to the Nth military code is,
Infantry division, infantry brigade, infantry battalion, infantry company, infantry platoon, infantry squad, reconnaissance battalion, reconnaissance company, reconnaissance platoon, artillery brigade, artillery battalion, artillery company, artillery unit, self-propelled artillery battalion, self-propelled artillery company, self-propelled artillery unit, Armored brigade, tank battalion, tank company, tank platoon, tank squad, mechanized infantry platoon, mechanized infantry squad, mechanized infantry division, mechanized infantry brigade, mechanized infantry battalion, mechanized infantry company, engineer brigade, engineer battalion, engineer company, engineer platoon , engineer squad, information and communications brigade, information and communications battalion, information and communications company, communications platoon, communications squad, aviation brigade, aviation battalion, aviation company, CBRN brigade, CBRN battalion, CBRN company, corps, attack helicopter battalion, medical corps, medical A military code recognition system characterized in that it is a symbol representing one of the following: a battalion, an air defense platoon, a mortar platoon, a large node communication station, a small node communication station, a unit communication station, and a relay station.
According to paragraph 1,
The artificial neural network for military code classification is,
A first network that detects the widest closed curve in the shape of a square or circle in the input military code image;
a second network that classifies a combination of shapes included in an inner area of a closed curve detected by the first network among the military code images;
a third network that classifies a combination of shapes included in an area outside the closed curve detected by the first network among the military code images; and
A military unit including a fourth network that outputs information to determine whether the military code image corresponds to one of the first military code and the Nth military code based on the result values output from the second network and the third network. Sign recognition system.
According to paragraph 1,
A first learning module for learning the artificial neural network for military code classification; and
It further includes a second learning module for learning the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle,
The first learning module is,
Learning an artificial neural network for classifying the military code by inputting a plurality of images for training a military code classification model tagged with information expressing which category it corresponds to among the 1st military code and the Nth military code,
Image augmentation is performed on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and the generated military code classification model additional training images are input to the military code classification model. Further learning an artificial neural network for code classification - Here, each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation,
The second learning module is for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training image includes the plurality of military code classification model training images and the plurality of military code classification model additional training images. Includes-,
Rotating the rotation angle classification model training image by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training image - each of the plurality of rotated images is tagged with a rotation angle - A military code recognition system that inputs a plurality of rotated images and learns an artificial neural network for determining angles corresponding to the images for learning the rotation angle classification model.
According to paragraph 1,
A military code recognition system, wherein the judgment target image is a handwritten image drawn by a user using a predetermined input device.
According to clause 5,
The military code recognition system is,
a map display module that displays a map image on a predetermined display device;
a UI providing module that provides a drawing UI for drawing the handwriting image;
a generation module that generates an object image obtained by rotating the kth standard military code by the rotation angle; and
A military code recognition system further comprising an object display module that displays the object image at a specific point on the map image selected by the user.
According to clause 6,
The acquisition module is,
Periodically acquire the judgment target image while the user is drawing on the drawing UI,
A military code recognition system that acquires the judgment target image whenever a line drawn by the user on the drawing UI is disconnected.
An artificial neural network for military code classification that receives a military code image and outputs information to determine whether the military code image corresponds to a 1st military code or an Nth military code (N is an integer of 2 or more); and
Artificial neural network for determining the first angle or artificial neural network for determining the N-th angle - Here, the artificial neural network for determining the i-th angle (i is an arbitrary integer with 1<=i<=N) is connected to the i-th group sign. It is a neural network that receives the image of the i military code and outputs information to determine how much the input image of the i standard military code is rotated compared to the predetermined i standard military code. module;
A first learning module for learning the artificial neural network for military code classification; and
A second learning module for learning the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle,
The first learning module is,
Learning an artificial neural network for classifying the military code by inputting a plurality of images for training a military code classification model tagged with information expressing which category it corresponds to among the 1st military code and the Nth military code,
Image augmentation is performed on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and the generated military code classification model additional training images are input to the military code classification model. Further learning an artificial neural network for code classification - Here, each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation,
The second learning module is for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training image includes the plurality of military code classification model training images and the plurality of military code classification model additional training images. Includes-,
Rotating the rotation angle classification model training image by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training image - each of the plurality of rotated images is tagged with a rotation angle - A neural network learning system for military code recognition that inputs a plurality of rotated images and learns an artificial neural network for determining angles corresponding to the images for learning the rotation angle classification model.
An artificial neural network for military code classification in which a computing system receives a military code image and outputs information to determine whether the military code image corresponds to the 1st military code or the Nth military code (N is an integer of 2 or more). ; and
Artificial neural network for determining the first angle or artificial neural network for determining the N-th angle - Here, the artificial neural network for determining the i-th angle (i is an arbitrary integer with 1<=i<=N) is connected to the i-th group sign. It is a neural network that receives the image of the i military code and outputs information to determine how much the input image of the i standard military code is rotated compared to the predetermined i standard military code. ;
Obtaining, by the computing system, an image to be judged;
The computing system inputs the judgment target image into the artificial neural network for military code classification, and determines the military code corresponding to the judgment target image based on the value output by the artificial neural network for military code classification. ; and
If it is determined that the judgment target image corresponds to the kth military code (k is any of the integers 1<=k<=N), the computing system sends the judgment target image to the kth angle judgment neural network. A military code recognition method comprising the step of inputting and determining a rotation angle indicating how much the judgment target image is rotated compared to the kth standard military code based on the value output by the neural network for determining the kth angle. .
According to clause 9,
The artificial neural network for military code classification is,
A first network that detects the widest closed curve in the shape of a square or circle in the input military code image;
a second network that classifies a combination of shapes included in an inner area of a closed curve detected by the first network among the military code images;
a third network that classifies a combination of shapes included in an area outside the closed curve detected by the first network among the military code images; and
A military unit including a fourth network that outputs information to determine whether the military code image corresponds to one of the first military code and the Nth military code based on the result values output from the second network and the third network. How to recognize signs.
According to clause 9,
learning, by the computing system, the artificial neural network for military code classification; and
The computing system further includes learning the artificial neural network for determining the first angle or the artificial neural network for determining the Nth angle,
The step of learning the artificial neural network for military code classification is,
Learning an artificial neural network for military code classification by inputting a plurality of images for training a military code classification model tagged with information representing which classification among the 1st military code and the Nth military code it corresponds to;
Image augmentation is performed on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and the generated military code classification model additional training images are input to the military code classification model. A step of further learning an artificial neural network for code classification - where each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation,
The step of learning the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle includes, for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training images are the plurality of rotation angle classification model training images. Includes images for military code classification model training and images for additional training of the multiple military code classification models -,
Rotating the rotation angle classification model training image by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training image - each of the plurality of rotated images is tagged with a rotation angle - A military code recognition method comprising the step of inputting a plurality of rotated images and learning an artificial neural network for determining angles corresponding to the images for learning the rotation angle classification model.
According to clause 9,
A military code recognition method, wherein the judgment target image is a handwritten image drawn by a user using a predetermined input device.
According to clause 12,
The military code recognition method is,
displaying, by the computing system, a map image on a predetermined display device;
providing, by the computing system, a drawing UI for drawing the handwriting image;
generating, by the computing system, an object image obtained by rotating the kth standard military code by the rotation angle; and
A military code recognition method further comprising the step of displaying, by the computing system, the object image at a specific point on the map image selected by the user.
According to clause 13,
The step of acquiring the judgment target image is,
periodically acquiring the judgment target image while the user is drawing on the drawing UI; or
A military code recognition method comprising acquiring the judgment target image whenever a line drawn by the user on the drawing UI is disconnected.
An artificial neural network for military code classification in which a computing system receives a military code image and outputs information to determine whether the military code image corresponds to the 1st military code or the Nth military code (N is an integer of 2 or more). ; and
The computing system includes an artificial neural network for determining a first angle or an artificial neural network for determining an N-th angle, where the artificial neural network for determining an i-th angle (i is an arbitrary integer where 1<=i<=N), It corresponds to the i military code, and is a neural network that receives the image of the i military code and outputs information to determine how much the input image of the i military code is rotated compared to the predetermined i standard military code. -Save it;
learning, by the computing system, the artificial neural network for military code classification; and
The computing system includes learning the artificial neural network for determining the first angle or the artificial neural network for determining the Nth angle,
The step of learning the artificial neural network for military code classification is,
Learning an artificial neural network for military code classification by inputting a plurality of images for training a military code classification model tagged with information representing which classification among the 1st military code and the Nth military code it corresponds to; and
Image augmentation is performed on each of the plurality of military code classification model training images to generate a plurality of military code classification model additional training images, and the generated military code classification model additional training images are input to the military code classification model. Further learning an artificial neural network for code classification - where each image for additional training of the military code classification model is tagged with the label of the original image before image augmentation,
The step of learning the artificial neural network for determining the first angle or the artificial neural network for determining the N-th angle includes, for each of a plurality of rotation angle classification model training images - where the rotation angle classification model training images are the plurality of rotation angle classification model training images. Includes images for military code classification model training and images for additional training of the multiple military code classification models -,
Rotating the rotation angle classification model training image by a predetermined unit angle to generate a plurality of rotated images corresponding to the rotation angle classification model training image - each of the plurality of rotated images is tagged with a rotation angle - A neural network learning method for military code recognition, comprising the step of inputting a plurality of rotated images and learning an artificial neural network for determining angles corresponding to the images for learning the rotation angle classification model.
A computer program installed in a data processing device and recorded on a medium for performing the method according to any one of claims 9 to 15.
A computer-readable recording medium on which a computer program for performing the method according to any one of claims 9 to 15 is recorded.
As a computing system,
processor; and a memory for storing a computer program,
The computer program, when executed by the processor, causes the computing system to perform the method according to any one of claims 9 to 15.